MozillaZine


Odd behavior on 2.53.6b1pre (for awhile)

Discussion about Seamonkey builds
RDaneel
 
Posts: 507
Joined: January 19th, 2004, 2:43 pm
Location: Seattle, WA

Post Posted December 7th, 2020, 2:57 pm

So, for the last couple of weeks, I have been seeing a "cluster" of behaviors that a) are all bad, and b) typically occur together... this is on my Win 10 Pro x64 box, scads of memory available.

My workflow and usage hasn't changed, typically just reading the historical Doonesbury strips for any given day, look at the Bing front page for news, check my orders on Amazon, keep up to date with a few GitHub projects (mot mine, in this case) - and a happy user of both "uBlock Origin" and NoScript. Presumably the most "off the norm" thing about my SM usage is that I keep ~200 tabs. ;)

For a few weeks now, I see the following: I notice while doing normal browsing that one of my Hyper-thread "cores" goes to being maxed out (12.5%, since I have 4 real cores, etc). Once I see this, I know I am hosed... what comes next is, a shutdown will not complete (and the CPU usage will often double to 25%, but not always), and I will not have my saved tabs (even though I said, yes, please save them).

For at least half-a-dozen builds this has persisted, the latest is on the build from yesterday 20201206... is anyone seeing all or part of this set of behaviors?

ndebord

User avatar
 
Posts: 922
Joined: December 7th, 2002, 9:53 am

Post Posted December 8th, 2020, 4:06 pm

RDaneel wrote:So, for the last couple of weeks, I have been seeing a "cluster" of behaviors that a) are all bad, and b) typically occur together... this is on my Win 10 Pro x64 box, scads of memory available.

For a few weeks now, I see the following: I notice while doing normal browsing that one of my Hyper-thread "cores" goes to being maxed out (12.5%, since I have 4 real cores, etc). Once I see this, I know I am hosed... what comes next is, a shutdown will not complete (and the CPU usage will often double to 25%, but not always), and I will not have my saved tabs (even though I said, yes, please save them).

For at least half-a-dozen builds this has persisted, the latest is on the build from yesterday 20201206... is anyone seeing all or part of this set of behaviors?


Not sure... don't follow which of the 4 cores I have on this Acer Spin1, however, I have been seeing recurring crashes since using .6 betas the last week.

Nick
-N- Quis custodiet ipsos custodes
SeaMonkey(32bit), Acer Spin, Windows 10 Pro (X64 v1909), WinPatrol, MalwarebytesPremium & Avira

RDaneel
 
Posts: 507
Joined: January 19th, 2004, 2:43 pm
Location: Seattle, WA

Post Posted December 9th, 2020, 2:31 pm

Thanks for the response, Nick, but to be clear, I am not talking about "crashes" per se - this is kind of a zombie / "walking crash" I am referring to. ;)
When I see this behavior with the obvious looping and sucking up an entire pseudo-core, I know that the shutdown will not be clean... until I want to do that shutdown, things are acting more or less normally.
But once I try the shutdown, I will see

a) SM will not shutdown,
b) SM will probably start using a second core, and
c) SM will continue its slow march of growing memory utilization

... and this is happening again right now, even though I had walked away from my system for an hour, leaving it sitting on an innocuous site (Bill's in this case). Sigh.

ndebord

User avatar
 
Posts: 922
Joined: December 7th, 2002, 9:53 am

Post Posted December 9th, 2020, 10:51 pm

RDaneel wrote:Thanks for the response, Nick, but to be clear, I am not talking about "crashes" per se - this is kind of a zombie / "walking crash" I am referring to. ;)
When I see this behavior with the obvious looping and sucking up an entire pseudo-core, I know that the shutdown will not be clean... until I want to do that shutdown, things are acting more or less normally.
But once I try the shutdown, I will see

a) SM will not shutdown,
b) SM will probably start using a second core, and
c) SM will continue its slow march of growing memory utilization

... and this is happening again right now, even though I had walked away from my system for an hour, leaving it sitting on an innocuous site (Bill's in this case). Sigh.


RDaneel

Well, it was getting to the point of too many crashes, so I went back to the official release and put the betas aside for now. I shall use System Explorer to check my cores to see if there is any similar behavior on my little laptop.

Nick
-N- Quis custodiet ipsos custodes
SeaMonkey(32bit), Acer Spin, Windows 10 Pro (X64 v1909), WinPatrol, MalwarebytesPremium & Avira

Snake4
 
Posts: 1051
Joined: December 27th, 2017, 4:03 am
Location: Australia

Post Posted December 9th, 2020, 11:17 pm

just a question, did you guys by any chance use a New Fresh Test Profile for this?

RDaneel
 
Posts: 507
Joined: January 19th, 2004, 2:43 pm
Location: Seattle, WA

Post Posted December 10th, 2020, 3:20 am

Indeed not (in my case) - that is why I am not "reporting a bug" in the sense of a specific single incorrect action that has a reasonably simple slimmed-down test case and is repeatable.

I have titled this and discuss what I am observing more as a "poll" to see if any others are in fact observing anything even remotely similar... if there were responses showing up that appeared to describe issues with some similarity to what I am seeing, then there might be enough of a pattern emerging that it would make sense to attempt to create an actual targeted test case.

Without even a vague suggestion of some known problem or issue that could possible relate to my observations, then I can only use SM as my daily driver "browser plus email client" in the way that I use it: with the 2 mentioned "protection" add-ons and with my ~200 tabs. ;)

Just stripping out these add-ons and running with a single tab (not to mention the empty profile) - and no notion of what I am even trying to isolate - what am I going to do, sit there and look at the thing? :p

To me, once one has at least a hypothesis regarding what observed issue at least might be getting caused by some underlying problem (aka "bug"), then simplified test cases start making sense - and I would then have some clue what I wanted to expose ("test") by setting up some specific situation.

Putting this another way, seeing strange things happening in real life usage and then building on that to create-a-targeted-hypothesis-and-test-it is one of the reasons - maybe the primary one - that we run "beta" software, right? :)

</ grumpy-and-tired-rant>

OK, one more topic / request: I saw a quick mention recently on the "freenode #seamonkey" discussion about the symbols files for SM builds... this might be fairly useful to at least occasionally publish, as in the problem I keep seeing, the running thread that starts sucking up all the CPU it can is always in the same sequence of calls (according to Process Hacker), so being able to attach to the SM process before I have to kill it - and make some minimal sense of where it might actually be and therefore what it might be doing [wrong] - could actually begin the process I talk about above.

ndebord

User avatar
 
Posts: 922
Joined: December 7th, 2002, 9:53 am

Post Posted December 10th, 2020, 8:10 am

Snake4 wrote:just a question, did you guys by any chance use a New Fresh Test Profile for this?



Snake4,

No, I keep a complete copy of SM on another drive, so don't bother with profile changes unless it seems I must do that.

Nick
-N- Quis custodiet ipsos custodes
SeaMonkey(32bit), Acer Spin, Windows 10 Pro (X64 v1909), WinPatrol, MalwarebytesPremium & Avira

RDaneel
 
Posts: 507
Joined: January 19th, 2004, 2:43 pm
Location: Seattle, WA

Post Posted December 13th, 2020, 12:47 pm

For those [still] playing along from home, an update...

For some unfounded reasons ultimately not worth exploring, I tried running "my usual" setup - but with "uBlock Origin" *disabled*. No problems on the latest Build From Bill(tm) from 20201212, but then I put Win 10 into "sleep" mode.

When I resumed today, as near as I can tell, I was instantly back in the mode of SM sucking up an entire HT "core". :(

i first looked from Process Hacker, and saw that the "usual suspect" thread was once again at the top... I attached to the process from VS, and had it "break all", and verified the super busy thread, and put together a picture using both VS and Process Hacker.

The thread's name is "Socket Thread", a "Normal" priority native thread, with the following at the top of the stack (before disappearing into system-land):

15, ntdll.dll!ZwDeviceIoControlFile+0x14
16, mswsock.dll+0x93bf
17, ws2_32.dll!send+0x16b
18, nss3.dll!PR_MD_UNLOCK+0x1831
19, nss3.dll!PR_GetLibraryFilePathname+0x22ed
20, nss3.dll!PR_GetLibraryFilePathname+0x1b24
21, xul.dll!Ordinal0+0x3e230a

And a bunch more XUL stack entries... this probably could be meaningful to someone that both a) worked on SM internals, and b) had the time to consider it.

Not sure the GetLibraryFilePathname entries mean anything in the themselves - with offsets that large, the code could be only distantly related - but it does look like SM is going off into the sockets stack for something...

Finally, I recalled something in the nature of a "pro tip": if I tell SM to shutdown and save all tabs, it says sure, but then of course - part of this issue - it doesn't actually complete the shutdown, but loses all the tabs. :(

However, if while it still thinks it is running normally (but actually already has this wild and looping thread), I do a "hard kill"-type terminate from Process Hacker, it does go away - but with all the tabs still intact. :)

frg
 
Posts: 1009
Joined: December 15th, 2015, 1:20 pm

Post Posted December 13th, 2020, 2:35 pm

Palemoon seems to have developed the same problem lately:

https://repo.palemoon.org/MoonchildProd ... ssues/1695

It does not occur here but I never but my computer to sleep either. Very rare I do it with the laptop but usually SeaMonkey is closed then.

FRG

RDaneel
 
Posts: 507
Joined: January 19th, 2004, 2:43 pm
Location: Seattle, WA

Post Posted December 13th, 2020, 9:47 pm

Thanks, f-r, I have now read that bug... I will say that in general, I have not seen a "tie" with this issue and sleep state being exited - I think it was just "lucky" that it happened so closely after resuming from sleep yesterday.

OTOH, I am using the polyfill thing for GitHub, and among my 200 tabs, yes, there are a couple of them looking at GH projects. ;)

So while I am willing to disable that add-in for now, and only re-enable it if I actually need the functionality it restores, this won't tell us all that much (just because we don't see the problem for a while, doesn't mean anything is actually gone etc)...

RDaneel
 
Posts: 507
Joined: January 19th, 2004, 2:43 pm
Location: Seattle, WA

Post Posted December 19th, 2020, 12:47 pm

Time for yet another update... it has now been 6 days running with the polyfill script disabled virtually all the time (see above).

So, as there is an updated version (-> 1.1.7, operative commit comment: "Update CSP workaround for socket-worker script"), that looks as if it might even be addressing the issue I was seeing, I have now updated the component from

https://github.com/JustOff/github-wc-polyfill/

... and will try running with it enabled all the time - and, of course, I have updated SM to Bill's build of 20201218.

Film @ 11 (as they say ;)).

RDaneel
 
Posts: 507
Joined: January 19th, 2004, 2:43 pm
Location: Seattle, WA

Post Posted January 3rd, 2021, 3:25 pm

Bummer... I was getting ready to post a closing "follow-up" on this, saying everything had been working fine - and just had a recurrence of the "loop sucking up a virtual core" behavior (I am running the 20210101 build of 2.53.7b1pre).

This was similar to some of the previous instances, where I had recently resumed from an overnight "sleep" of Windows 10 - with SeaMonkey experiencing that sleep, as in, not shut down first.

I just checked the site for the "GitHub Web Components Polyfill 1.1.7", and there hasn't been another release since then, so I have once again disabled that plug-in.

For all I know, it was fine and [possibly] GitHub has changed again, and done something else that they shouldn't be doing on their site. Sigh.

Return to SeaMonkey Builds


Who is online

Users browsing this forum: No registered users and 1 guest